Case study · B2B services: Email on autopilot: AI that sends, replies and follows up for a sales team

An AI email system that sends personalised outreach, reads and classifies every reply, drafts answers and schedules follow-ups, so the sales team only handles conversations that are ready to buy.

Client: A US B2B services company
Industry: B2B services
Country: United States
Year: 2025
Timeline: 10 weeks
Cloud: AWS

System architecture

How the pieces connect

AWS
Live flow

01Leads

HubSpot CRM
Approved content

02Send

AWS Step FunctionsAWS
Amazon SQSAWS
Amazon SESAWS

03Receive

SES inboundAWS
Amazon S3AWS
AWS LambdaAWS

04Understand

Amazon BedrockAWS
Amazon DynamoDBAWS

05Act

EventBridge SchedulerAWS
Rep approval
CRM sync

(01)Results

What changed.

30+ hrs

Manual email work removed each week

< 5 min

Response time to inbound replies

100%

Of leads followed up on schedule

(02)The challenge

The problem.

The client's sales team ran outreach from shared inboxes and spreadsheets. Each rep sent campaigns by hand, read every reply, decided what to do with it and set their own reminders for follow-ups.

As volume grew, replies sat unanswered for hours, follow-ups were missed when someone was busy or on leave, and nobody had a reliable view of which leads were interested. Good opportunities were going cold before anyone noticed them.

(03)The solution

What we built.

We built an event-driven email automation platform on AWS. It sends personalised sequences from verified domains, receives every reply, and uses an LLM to classify it as interested, question, objection, out of office, not now or unsubscribe.

Each category triggers its own workflow. Questions get a drafted answer grounded in the client's approved content, interested leads are routed to the right rep with a summary, 'not now' replies are rescheduled, and unsubscribes are honoured instantly. Follow-ups are timed automatically and stop the moment someone replies. Reps approve anything sensitive before it goes out.

Personalised sending

Sequences with per-lead personalisation, sending limits and domain warm-up to protect deliverability.

AI reply triage

Every inbound reply is read, classified and summarised within minutes of arriving.

Drafted answers

Answers drafted from approved content, sent automatically for routine cases and queued for approval otherwise.

Smart follow-ups

Timed follow-ups that pause on reply, respect time zones and reschedule 'not now' leads.

(04)Architecture

9 cloud services, one system.

AWS

Amazon SES
Outbound sending and inbound email receiving
Amazon S3
Storage of raw inbound emails and attachments
AWS Lambda
Parsing replies, personalisation and CRM sync
Amazon Bedrock
Reply classification, summaries and drafted answers
AWS Step Functions
Orchestrates each sequence and reply workflow
Amazon EventBridge Scheduler
Schedules follow-ups in each lead's time zone
Amazon SQS
Smooths sending bursts to stay within limits
Amazon DynamoDB
Leads, sequences, conversation state and events
Amazon CloudWatch
Bounce, complaint and delivery monitoring with alerts

Stack

Python
TypeScript
React
HubSpot API
AWS CDK

(05)The outcome

After launch.

The sales team stopped managing inboxes and started managing conversations. Routine replies are handled in minutes, every lead gets its follow-ups on time, and reps start each day with a short list of interested prospects and a summary of each thread.

Because every step is logged, the client can see exactly what was sent, why, and how each lead responded, which they now use to refine their messaging.

“We used to lose leads simply because nobody replied in time. Now every reply is handled within minutes and my team spends its day on real conversations.”
Head of Sales, US B2B services company

Services in this project